hugging-face-datasets

Run SQL queries on Hugging Face datasets using DuckDB-backed access to hf:// paths, enabling describe, sample, and export operations. This SQL-driven workflow lets data engineers explore and filter Hub data directly.

The best way to create a curated subset is to initialize a new repo, apply multi-format templates like chat or classification, query and filter with DuckDB SQL, then push the result as a new Hub repository for sharing.

Initialize a new Hugging Face dataset repo using template-driven validation for chat, classification, QA, completion, or tabular formats. This supports configuring system prompts and streaming sample rows directly to the Hub.

Yes, DuckDB provides SQL access to hf:// Hugging Face datasets and supports streaming updates to the Hub. It integrates with repo initialization and multi-format templates for dataset lifecycle management.

Supported formats include chat, classification, QA, completion, and tabular data. These multi-format templates include validation to ensure correct data structure when you push updates to the Hub.

Yes, the DuckDB-backed SQL access supports export operations alongside describe and sample commands. You can extract queried data from hf:// datasets for downstream analysis or push subsets as new Hub repositories.